CSR---EHS: A Framework for Optimizing Embedded Systems

University Of Arizona, Tucson AZ

Investigators

Abstract

Embedded systems have historically executed on primitive hardware and consequently have had limited functionality. Hardware developments have led to much more powerful platforms, and industrial and consumer demands have led to more powerful applications. This in turn has resulted in the delivery and use of systems that contain a full operating system kernel, such as Linux or one of its embedded systems variants. These trends seem certain to continue, but hardware resources will always remain limited relative to more powerful machines because embedded systems must usually be small and inexpensive and often are battery operated. This project aims to assist the development of future embedded systems by automating the optimization process, thus allowing system developers to focus on functionality and spend much less time worrying about resource constraints such as limited memory or power. In particular, the project is creating a reusable optimization framework that takes an entire embedded system (applications, library programs, and operating system kernel) and uses profile-guided optimization to produce a new version of the system that uses much less memory, executes faster, and uses less energy. To realize these goals, the project is developing techniques that, given an embedded system with a fixed set of applications: (1) ensure that we have complete knowledge of application/kernel interactions (using a combination of profiling and static analysis), (2) optimize the system to reduce code size, execution time, and energy usage (using binary rewriting), and (3) ensure that the resulting system can be understood, tested, and maintained.